arch/s390/kvm/vsie.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * kvm nested virtualization support for s390x
   4  *
   5  * Copyright IBM Corp. 2016, 2018
   6  *
   7  *    Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com>
   8  */
   9 #include <linux/vmalloc.h>
  10 #include <linux/kvm_host.h>
  11 #include <linux/bug.h>
  12 #include <linux/list.h>
  13 #include <linux/bitmap.h>
  14 #include <linux/sched/signal.h>
  15
  16 #include <asm/gmap.h>
  17 #include <asm/mmu_context.h>
  18 #include <asm/sclp.h>
  19 #include <asm/nmi.h>
  20 #include <asm/dis.h>
  21 #include "kvm-s390.h"
  22 #include "gaccess.h"
  23
  24 struct vsie_page {
  25         struct kvm_s390_sie_block scb_s;        /* 0x0000 */
  26         /*
  27          * the backup info for machine check. ensure it's at
  28          * the same offset as that in struct sie_page!
  29          */
  30         struct mcck_volatile_info mcck_info;    /* 0x0200 */
  31         /*
  32          * The pinned original scb. Be aware that other VCPUs can modify
  33          * it while we read from it. Values that are used for conditions or
  34          * are reused conditionally, should be accessed via READ_ONCE.
  35          */
  36         struct kvm_s390_sie_block *scb_o;       /* 0x0218 */
  37         /* the shadow gmap in use by the vsie_page */
  38         struct gmap *gmap;                      /* 0x0220 */
  39         /* address of the last reported fault to guest2 */
  40         unsigned long fault_addr;               /* 0x0228 */
  41         /* calculated guest addresses of satellite control blocks */
  42         gpa_t sca_gpa;                          /* 0x0230 */
  43         gpa_t itdba_gpa;                        /* 0x0238 */
  44         gpa_t gvrd_gpa;                         /* 0x0240 */
  45         gpa_t riccbd_gpa;                       /* 0x0248 */
  46         gpa_t sdnx_gpa;                         /* 0x0250 */
  47         __u8 reserved[0x0700 - 0x0258];         /* 0x0258 */
  48         struct kvm_s390_crypto_cb crycb;        /* 0x0700 */
  49         __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
  50 };
  51
  52 /* trigger a validity icpt for the given scb */
  53 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
  54                              __u16 reason_code)
  55 {
  56         scb->ipa = 0x1000;
  57         scb->ipb = ((__u32) reason_code) << 16;
  58         scb->icptcode = ICPT_VALIDITY;
  59         return 1;
  60 }
  61
  62 /* mark the prefix as unmapped, this will block the VSIE */
  63 static void prefix_unmapped(struct vsie_page *vsie_page)
  64 {
  65         atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
  66 }
  67
  68 /* mark the prefix as unmapped and wait until the VSIE has been left */
  69 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
  70 {
  71         prefix_unmapped(vsie_page);
  72         if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
  73                 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
  74         while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
  75                 cpu_relax();
  76 }
  77
  78 /* mark the prefix as mapped, this will allow the VSIE to run */
  79 static void prefix_mapped(struct vsie_page *vsie_page)
  80 {
  81         atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
  82 }
  83
  84 /* test if the prefix is mapped into the gmap shadow */
  85 static int prefix_is_mapped(struct vsie_page *vsie_page)
  86 {
  87         return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
  88 }
  89
  90 /* copy the updated intervention request bits into the shadow scb */
  91 static void update_intervention_requests(struct vsie_page *vsie_page)
  92 {
  93         const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
  94         int cpuflags;
  95
  96         cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
  97         atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
  98         atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
  99 }
 100
 101 /* shadow (filter and validate) the cpuflags  */
 102 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 103 {
 104         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 105         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 106         int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
 107
 108         /* we don't allow ESA/390 guests */
 109         if (!(cpuflags & CPUSTAT_ZARCH))
 110                 return set_validity_icpt(scb_s, 0x0001U);
 111
 112         if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
 113                 return set_validity_icpt(scb_s, 0x0001U);
 114         else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
 115                 return set_validity_icpt(scb_s, 0x0007U);
 116
 117         /* intervention requests will be set later */
 118         newflags = CPUSTAT_ZARCH;
 119         if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
 120                 newflags |= CPUSTAT_GED;
 121         if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
 122                 if (cpuflags & CPUSTAT_GED)
 123                         return set_validity_icpt(scb_s, 0x0001U);
 124                 newflags |= CPUSTAT_GED2;
 125         }
 126         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
 127                 newflags |= cpuflags & CPUSTAT_P;
 128         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
 129                 newflags |= cpuflags & CPUSTAT_SM;
 130         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
 131                 newflags |= cpuflags & CPUSTAT_IBS;
 132         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
 133                 newflags |= cpuflags & CPUSTAT_KSS;
 134
 135         atomic_set(&scb_s->cpuflags, newflags);
 136         return 0;
 137 }
 138 /* Copy to APCB FORMAT1 from APCB FORMAT0 */
 139 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
 140                         unsigned long apcb_o, struct kvm_s390_apcb1 *apcb_h)
 141 {
 142         struct kvm_s390_apcb0 tmp;
 143
 144         if (read_guest_real(vcpu, apcb_o, &tmp, sizeof(struct kvm_s390_apcb0)))
 145                 return -EFAULT;
 146
 147         apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
 148         apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
 149         apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
 150
 151         return 0;
 152
 153 }
 154
 155 /**
 156  * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
 157  * @vcpu: pointer to the virtual CPU
 158  * @apcb_s: pointer to start of apcb in the shadow crycb
 159  * @apcb_o: pointer to start of original apcb in the guest2
 160  * @apcb_h: pointer to start of apcb in the guest1
 161  *
 162  * Returns 0 and -EFAULT on error reading guest apcb
 163  */
 164 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 165                         unsigned long apcb_o, unsigned long *apcb_h)
 166 {
 167         if (read_guest_real(vcpu, apcb_o, apcb_s,
 168                             sizeof(struct kvm_s390_apcb0)))
 169                 return -EFAULT;
 170
 171         bitmap_and(apcb_s, apcb_s, apcb_h,
 172                    BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0));
 173
 174         return 0;
 175 }
 176
 177 /**
 178  * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
 179  * @vcpu: pointer to the virtual CPU
 180  * @apcb_s: pointer to start of apcb in the shadow crycb
 181  * @apcb_o: pointer to start of original guest apcb
 182  * @apcb_h: pointer to start of apcb in the host
 183  *
 184  * Returns 0 and -EFAULT on error reading guest apcb
 185  */
 186 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 187                         unsigned long apcb_o,
 188                         unsigned long *apcb_h)
 189 {
 190         if (read_guest_real(vcpu, apcb_o, apcb_s,
 191                             sizeof(struct kvm_s390_apcb1)))
 192                 return -EFAULT;
 193
 194         bitmap_and(apcb_s, apcb_s, apcb_h,
 195                    BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1));
 196
 197         return 0;
 198 }
 199
 200 /**
 201  * setup_apcb - Create a shadow copy of the apcb.
 202  * @vcpu: pointer to the virtual CPU
 203  * @crycb_s: pointer to shadow crycb
 204  * @crycb_o: pointer to original guest crycb
 205  * @crycb_h: pointer to the host crycb
 206  * @fmt_o: format of the original guest crycb.
 207  * @fmt_h: format of the host crycb.
 208  *
 209  * Checks the compatibility between the guest and host crycb and calls the
 210  * appropriate copy function.
 211  *
 212  * Return 0 or an error number if the guest and host crycb are incompatible.
 213  */
 214 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
 215                const u32 crycb_o,
 216                struct kvm_s390_crypto_cb *crycb_h,
 217                int fmt_o, int fmt_h)
 218 {
 219         struct kvm_s390_crypto_cb *crycb;
 220
 221         crycb = (struct kvm_s390_crypto_cb *) (unsigned long)crycb_o;
 222
 223         switch (fmt_o) {
 224         case CRYCB_FORMAT2:
 225                 if ((crycb_o & PAGE_MASK) != ((crycb_o + 256) & PAGE_MASK))
 226                         return -EACCES;
 227                 if (fmt_h != CRYCB_FORMAT2)
 228                         return -EINVAL;
 229                 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
 230                                     (unsigned long) &crycb->apcb1,
 231                                     (unsigned long *)&crycb_h->apcb1);
 232         case CRYCB_FORMAT1:
 233                 switch (fmt_h) {
 234                 case CRYCB_FORMAT2:
 235                         return setup_apcb10(vcpu, &crycb_s->apcb1,
 236                                             (unsigned long) &crycb->apcb0,
 237                                             &crycb_h->apcb1);
 238                 case CRYCB_FORMAT1:
 239                         return setup_apcb00(vcpu,
 240                                             (unsigned long *) &crycb_s->apcb0,
 241                                             (unsigned long) &crycb->apcb0,
 242                                             (unsigned long *) &crycb_h->apcb0);
 243                 }
 244                 break;
 245         case CRYCB_FORMAT0:
 246                 if ((crycb_o & PAGE_MASK) != ((crycb_o + 32) & PAGE_MASK))
 247                         return -EACCES;
 248
 249                 switch (fmt_h) {
 250                 case CRYCB_FORMAT2:
 251                         return setup_apcb10(vcpu, &crycb_s->apcb1,
 252                                             (unsigned long) &crycb->apcb0,
 253                                             &crycb_h->apcb1);
 254                 case CRYCB_FORMAT1:
 255                 case CRYCB_FORMAT0:
 256                         return setup_apcb00(vcpu,
 257                                             (unsigned long *) &crycb_s->apcb0,
 258                                             (unsigned long) &crycb->apcb0,
 259                                             (unsigned long *) &crycb_h->apcb0);
 260                 }
 261         }
 262         return -EINVAL;
 263 }
 264
 265 /**
 266  * shadow_crycb - Create a shadow copy of the crycb block
 267  * @vcpu: a pointer to the virtual CPU
 268  * @vsie_page: a pointer to internal date used for the vSIE
 269  *
 270  * Create a shadow copy of the crycb block and setup key wrapping, if
 271  * requested for guest 3 and enabled for guest 2.
 272  *
 273  * We accept format-1 or format-2, but we convert format-1 into format-2
 274  * in the shadow CRYCB.
 275  * Using format-2 enables the firmware to choose the right format when
 276  * scheduling the SIE.
 277  * There is nothing to do for format-0.
 278  *
 279  * This function centralize the issuing of set_validity_icpt() for all
 280  * the subfunctions working on the crycb.
 281  *
 282  * Returns: - 0 if shadowed or nothing to do
 283  *          - > 0 if control has to be given to guest 2
 284  */
 285 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 286 {
 287         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 288         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 289         const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
 290         const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
 291         unsigned long *b1, *b2;
 292         u8 ecb3_flags;
 293         u32 ecd_flags;
 294         int apie_h;
 295         int apie_s;
 296         int key_msk = test_kvm_facility(vcpu->kvm, 76);
 297         int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
 298         int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
 299         int ret = 0;
 300
 301         scb_s->crycbd = 0;
 302
 303         apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
 304         apie_s = apie_h & scb_o->eca;
 305         if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
 306                 return 0;
 307
 308         if (!crycb_addr)
 309                 return set_validity_icpt(scb_s, 0x0039U);
 310
 311         if (fmt_o == CRYCB_FORMAT1)
 312                 if ((crycb_addr & PAGE_MASK) !=
 313                     ((crycb_addr + 128) & PAGE_MASK))
 314                         return set_validity_icpt(scb_s, 0x003CU);
 315
 316         if (apie_s) {
 317                 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
 318                                  vcpu->kvm->arch.crypto.crycb,
 319                                  fmt_o, fmt_h);
 320                 if (ret)
 321                         goto end;
 322                 scb_s->eca |= scb_o->eca & ECA_APIE;
 323         }
 324
 325         /* we may only allow it if enabled for guest 2 */
 326         ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
 327                      (ECB3_AES | ECB3_DEA);
 328         ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC;
 329         if (!ecb3_flags && !ecd_flags)
 330                 goto end;
 331
 332         /* copy only the wrapping keys */
 333         if (read_guest_real(vcpu, crycb_addr + 72,
 334                             vsie_page->crycb.dea_wrapping_key_mask, 56))
 335                 return set_validity_icpt(scb_s, 0x0035U);
 336
 337         scb_s->ecb3 |= ecb3_flags;
 338         scb_s->ecd |= ecd_flags;
 339
 340         /* xor both blocks in one run */
 341         b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
 342         b2 = (unsigned long *)
 343                             vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
 344         /* as 56%8 == 0, bitmap_xor won't overwrite any data */
 345         bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
 346 end:
 347         switch (ret) {
 348         case -EINVAL:
 349                 return set_validity_icpt(scb_s, 0x0022U);
 350         case -EFAULT:
 351                 return set_validity_icpt(scb_s, 0x0035U);
 352         case -EACCES:
 353                 return set_validity_icpt(scb_s, 0x003CU);
 354         }
 355         scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2;
 356         return 0;
 357 }
 358
 359 /* shadow (round up/down) the ibc to avoid validity icpt */
 360 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 361 {
 362         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 363         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 364         /* READ_ONCE does not work on bitfields - use a temporary variable */
 365         const uint32_t __new_ibc = scb_o->ibc;
 366         const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
 367         __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
 368
 369         scb_s->ibc = 0;
 370         /* ibc installed in g2 and requested for g3 */
 371         if (vcpu->kvm->arch.model.ibc && new_ibc) {
 372                 scb_s->ibc = new_ibc;
 373                 /* takte care of the minimum ibc level of the machine */
 374                 if (scb_s->ibc < min_ibc)
 375                         scb_s->ibc = min_ibc;
 376                 /* take care of the maximum ibc level set for the guest */
 377                 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
 378                         scb_s->ibc = vcpu->kvm->arch.model.ibc;
 379         }
 380 }
 381
 382 /* unshadow the scb, copying parameters back to the real scb */
 383 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 384 {
 385         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 386         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 387
 388         /* interception */
 389         scb_o->icptcode = scb_s->icptcode;
 390         scb_o->icptstatus = scb_s->icptstatus;
 391         scb_o->ipa = scb_s->ipa;
 392         scb_o->ipb = scb_s->ipb;
 393         scb_o->gbea = scb_s->gbea;
 394
 395         /* timer */
 396         scb_o->cputm = scb_s->cputm;
 397         scb_o->ckc = scb_s->ckc;
 398         scb_o->todpr = scb_s->todpr;
 399
 400         /* guest state */
 401         scb_o->gpsw = scb_s->gpsw;
 402         scb_o->gg14 = scb_s->gg14;
 403         scb_o->gg15 = scb_s->gg15;
 404         memcpy(scb_o->gcr, scb_s->gcr, 128);
 405         scb_o->pp = scb_s->pp;
 406
 407         /* branch prediction */
 408         if (test_kvm_facility(vcpu->kvm, 82)) {
 409                 scb_o->fpf &= ~FPF_BPBC;
 410                 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
 411         }
 412
 413         /* interrupt intercept */
 414         switch (scb_s->icptcode) {
 415         case ICPT_PROGI:
 416         case ICPT_INSTPROGI:
 417         case ICPT_EXTINT:
 418                 memcpy((void *)((u64)scb_o + 0xc0),
 419                        (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
 420                 break;
 421         case ICPT_PARTEXEC:
 422                 /* MVPG only */
 423                 memcpy((void *)((u64)scb_o + 0xc0),
 424                        (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
 425                 break;
 426         }
 427
 428         if (scb_s->ihcpu != 0xffffU)
 429                 scb_o->ihcpu = scb_s->ihcpu;
 430 }
 431
 432 /*
 433  * Setup the shadow scb by copying and checking the relevant parts of the g2
 434  * provided scb.
 435  *
 436  * Returns: - 0 if the scb has been shadowed
 437  *          - > 0 if control has to be given to guest 2
 438  */
 439 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 440 {
 441         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 442         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 443         /* READ_ONCE does not work on bitfields - use a temporary variable */
 444         const uint32_t __new_prefix = scb_o->prefix;
 445         const uint32_t new_prefix = READ_ONCE(__new_prefix);
 446         const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
 447         bool had_tx = scb_s->ecb & ECB_TE;
 448         unsigned long new_mso = 0;
 449         int rc;
 450
 451         /* make sure we don't have any leftovers when reusing the scb */
 452         scb_s->icptcode = 0;
 453         scb_s->eca = 0;
 454         scb_s->ecb = 0;
 455         scb_s->ecb2 = 0;
 456         scb_s->ecb3 = 0;
 457         scb_s->ecd = 0;
 458         scb_s->fac = 0;
 459         scb_s->fpf = 0;
 460
 461         rc = prepare_cpuflags(vcpu, vsie_page);
 462         if (rc)
 463                 goto out;
 464
 465         /* timer */
 466         scb_s->cputm = scb_o->cputm;
 467         scb_s->ckc = scb_o->ckc;
 468         scb_s->todpr = scb_o->todpr;
 469         scb_s->epoch = scb_o->epoch;
 470
 471         /* guest state */
 472         scb_s->gpsw = scb_o->gpsw;
 473         scb_s->gg14 = scb_o->gg14;
 474         scb_s->gg15 = scb_o->gg15;
 475         memcpy(scb_s->gcr, scb_o->gcr, 128);
 476         scb_s->pp = scb_o->pp;
 477
 478         /* interception / execution handling */
 479         scb_s->gbea = scb_o->gbea;
 480         scb_s->lctl = scb_o->lctl;
 481         scb_s->svcc = scb_o->svcc;
 482         scb_s->ictl = scb_o->ictl;
 483         /*
 484          * SKEY handling functions can't deal with false setting of PTE invalid
 485          * bits. Therefore we cannot provide interpretation and would later
 486          * have to provide own emulation handlers.
 487          */
 488         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
 489                 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 490
 491         scb_s->icpua = scb_o->icpua;
 492
 493         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
 494                 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
 495         /* if the hva of the prefix changes, we have to remap the prefix */
 496         if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
 497                 prefix_unmapped(vsie_page);
 498          /* SIE will do mso/msl validity and exception checks for us */
 499         scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
 500         scb_s->mso = new_mso;
 501         scb_s->prefix = new_prefix;
 502
 503         /* We have to definetly flush the tlb if this scb never ran */
 504         if (scb_s->ihcpu != 0xffffU)
 505                 scb_s->ihcpu = scb_o->ihcpu;
 506
 507         /* MVPG and Protection Exception Interpretation are always available */
 508         scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
 509         /* Host-protection-interruption introduced with ESOP */
 510         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
 511                 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
 512         /* transactional execution */
 513         if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
 514                 /* remap the prefix is tx is toggled on */
 515                 if (!had_tx)
 516                         prefix_unmapped(vsie_page);
 517                 scb_s->ecb |= ECB_TE;
 518         }
 519         /* branch prediction */
 520         if (test_kvm_facility(vcpu->kvm, 82))
 521                 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
 522         /* SIMD */
 523         if (test_kvm_facility(vcpu->kvm, 129)) {
 524                 scb_s->eca |= scb_o->eca & ECA_VX;
 525                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 526         }
 527         /* Run-time-Instrumentation */
 528         if (test_kvm_facility(vcpu->kvm, 64))
 529                 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
 530         /* Instruction Execution Prevention */
 531         if (test_kvm_facility(vcpu->kvm, 130))
 532                 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
 533         /* Guarded Storage */
 534         if (test_kvm_facility(vcpu->kvm, 133)) {
 535                 scb_s->ecb |= scb_o->ecb & ECB_GS;
 536                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 537         }
 538         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
 539                 scb_s->eca |= scb_o->eca & ECA_SII;
 540         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
 541                 scb_s->eca |= scb_o->eca & ECA_IB;
 542         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
 543                 scb_s->eca |= scb_o->eca & ECA_CEI;
 544         /* Epoch Extension */
 545         if (test_kvm_facility(vcpu->kvm, 139)) {
 546                 scb_s->ecd |= scb_o->ecd & ECD_MEF;
 547                 scb_s->epdx = scb_o->epdx;
 548         }
 549
 550         /* etoken */
 551         if (test_kvm_facility(vcpu->kvm, 156))
 552                 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
 553
 554         scb_s->hpid = HPID_VSIE;
 555
 556         prepare_ibc(vcpu, vsie_page);
 557         rc = shadow_crycb(vcpu, vsie_page);
 558 out:
 559         if (rc)
 560                 unshadow_scb(vcpu, vsie_page);
 561         return rc;
 562 }
 563
 564 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
 565                                  unsigned long end)
 566 {
 567         struct kvm *kvm = gmap->private;
 568         struct vsie_page *cur;
 569         unsigned long prefix;
 570         struct page *page;
 571         int i;
 572
 573         if (!gmap_is_shadow(gmap))
 574                 return;
 575         if (start >= 1UL << 31)
 576                 /* We are only interested in prefix pages */
 577                 return;
 578
 579         /*
 580          * Only new shadow blocks are added to the list during runtime,
 581          * therefore we can safely reference them all the time.
 582          */
 583         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
 584                 page = READ_ONCE(kvm->arch.vsie.pages[i]);
 585                 if (!page)
 586                         continue;
 587                 cur = page_to_virt(page);
 588                 if (READ_ONCE(cur->gmap) != gmap)
 589                         continue;
 590                 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
 591                 /* with mso/msl, the prefix lies at an offset */
 592                 prefix += cur->scb_s.mso;
 593                 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
 594                         prefix_unmapped_sync(cur);
 595         }
 596 }
 597
 598 /*
 599  * Map the first prefix page and if tx is enabled also the second prefix page.
 600  *
 601  * The prefix will be protected, a gmap notifier will inform about unmaps.
 602  * The shadow scb must not be executed until the prefix is remapped, this is
 603  * guaranteed by properly handling PROG_REQUEST.
 604  *
 605  * Returns: - 0 on if successfully mapped or already mapped
 606  *          - > 0 if control has to be given to guest 2
 607  *          - -EAGAIN if the caller can retry immediately
 608  *          - -ENOMEM if out of memory
 609  */
 610 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 611 {
 612         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 613         u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
 614         int rc;
 615
 616         if (prefix_is_mapped(vsie_page))
 617                 return 0;
 618
 619         /* mark it as mapped so we can catch any concurrent unmappers */
 620         prefix_mapped(vsie_page);
 621
 622         /* with mso/msl, the prefix lies at offset *mso* */
 623         prefix += scb_s->mso;
 624
 625         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
 626         if (!rc && (scb_s->ecb & ECB_TE))
 627                 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 628                                            prefix + PAGE_SIZE);
 629         /*
 630          * We don't have to mprotect, we will be called for all unshadows.
 631          * SIE will detect if protection applies and trigger a validity.
 632          */
 633         if (rc)
 634                 prefix_unmapped(vsie_page);
 635         if (rc > 0 || rc == -EFAULT)
 636                 rc = set_validity_icpt(scb_s, 0x0037U);
 637         return rc;
 638 }
 639
 640 /*
 641  * Pin the guest page given by gpa and set hpa to the pinned host address.
 642  * Will always be pinned writable.
 643  *
 644  * Returns: - 0 on success
 645  *          - -EINVAL if the gpa is not valid guest storage
 646  */
 647 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
 648 {
 649         struct page *page;
 650
 651         page = gfn_to_page(kvm, gpa_to_gfn(gpa));
 652         if (is_error_page(page))
 653                 return -EINVAL;
 654         *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
 655         return 0;
 656 }
 657
 658 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
 659 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
 660 {
 661         kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
 662         /* mark the page always as dirty for migration */
 663         mark_page_dirty(kvm, gpa_to_gfn(gpa));
 664 }
 665
 666 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
 667 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 668 {
 669         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 670         hpa_t hpa;
 671
 672         hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
 673         if (hpa) {
 674                 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
 675                 vsie_page->sca_gpa = 0;
 676                 scb_s->scaol = 0;
 677                 scb_s->scaoh = 0;
 678         }
 679
 680         hpa = scb_s->itdba;
 681         if (hpa) {
 682                 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
 683                 vsie_page->itdba_gpa = 0;
 684                 scb_s->itdba = 0;
 685         }
 686
 687         hpa = scb_s->gvrd;
 688         if (hpa) {
 689                 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
 690                 vsie_page->gvrd_gpa = 0;
 691                 scb_s->gvrd = 0;
 692         }
 693
 694         hpa = scb_s->riccbd;
 695         if (hpa) {
 696                 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
 697                 vsie_page->riccbd_gpa = 0;
 698                 scb_s->riccbd = 0;
 699         }
 700
 701         hpa = scb_s->sdnxo;
 702         if (hpa) {
 703                 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
 704                 vsie_page->sdnx_gpa = 0;
 705                 scb_s->sdnxo = 0;
 706         }
 707 }
 708
 709 /*
 710  * Instead of shadowing some blocks, we can simply forward them because the
 711  * addresses in the scb are 64 bit long.
 712  *
 713  * This works as long as the data lies in one page. If blocks ever exceed one
 714  * page, we have to fall back to shadowing.
 715  *
 716  * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
 717  * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
 718  *
 719  * Returns: - 0 if all blocks were pinned.
 720  *          - > 0 if control has to be given to guest 2
 721  *          - -ENOMEM if out of memory
 722  */
 723 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 724 {
 725         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 726         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 727         hpa_t hpa;
 728         gpa_t gpa;
 729         int rc = 0;
 730
 731         gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
 732         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
 733                 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
 734         if (gpa) {
 735                 if (gpa < 2 * PAGE_SIZE)
 736                         rc = set_validity_icpt(scb_s, 0x0038U);
 737                 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
 738                         rc = set_validity_icpt(scb_s, 0x0011U);
 739                 else if ((gpa & PAGE_MASK) !=
 740                          ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
 741                         rc = set_validity_icpt(scb_s, 0x003bU);
 742                 if (!rc) {
 743                         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 744                         if (rc)
 745                                 rc = set_validity_icpt(scb_s, 0x0034U);
 746                 }
 747                 if (rc)
 748                         goto unpin;
 749                 vsie_page->sca_gpa = gpa;
 750                 scb_s->scaoh = (u32)((u64)hpa >> 32);
 751                 scb_s->scaol = (u32)(u64)hpa;
 752         }
 753
 754         gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
 755         if (gpa && (scb_s->ecb & ECB_TE)) {
 756                 if (gpa < 2 * PAGE_SIZE) {
 757                         rc = set_validity_icpt(scb_s, 0x0080U);
 758                         goto unpin;
 759                 }
 760                 /* 256 bytes cannot cross page boundaries */
 761                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 762                 if (rc) {
 763                         rc = set_validity_icpt(scb_s, 0x0080U);
 764                         goto unpin;
 765                 }
 766                 vsie_page->itdba_gpa = gpa;
 767                 scb_s->itdba = hpa;
 768         }
 769
 770         gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
 771         if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
 772                 if (gpa < 2 * PAGE_SIZE) {
 773                         rc = set_validity_icpt(scb_s, 0x1310U);
 774                         goto unpin;
 775                 }
 776                 /*
 777                  * 512 bytes vector registers cannot cross page boundaries
 778                  * if this block gets bigger, we have to shadow it.
 779                  */
 780                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 781                 if (rc) {
 782                         rc = set_validity_icpt(scb_s, 0x1310U);
 783                         goto unpin;
 784                 }
 785                 vsie_page->gvrd_gpa = gpa;
 786                 scb_s->gvrd = hpa;
 787         }
 788
 789         gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
 790         if (gpa && (scb_s->ecb3 & ECB3_RI)) {
 791                 if (gpa < 2 * PAGE_SIZE) {
 792                         rc = set_validity_icpt(scb_s, 0x0043U);
 793                         goto unpin;
 794                 }
 795                 /* 64 bytes cannot cross page boundaries */
 796                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 797                 if (rc) {
 798                         rc = set_validity_icpt(scb_s, 0x0043U);
 799                         goto unpin;
 800                 }
 801                 /* Validity 0x0044 will be checked by SIE */
 802                 vsie_page->riccbd_gpa = gpa;
 803                 scb_s->riccbd = hpa;
 804         }
 805         if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
 806             (scb_s->ecd & ECD_ETOKENF)) {
 807                 unsigned long sdnxc;
 808
 809                 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
 810                 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
 811                 if (!gpa || gpa < 2 * PAGE_SIZE) {
 812                         rc = set_validity_icpt(scb_s, 0x10b0U);
 813                         goto unpin;
 814                 }
 815                 if (sdnxc < 6 || sdnxc > 12) {
 816                         rc = set_validity_icpt(scb_s, 0x10b1U);
 817                         goto unpin;
 818                 }
 819                 if (gpa & ((1 << sdnxc) - 1)) {
 820                         rc = set_validity_icpt(scb_s, 0x10b2U);
 821                         goto unpin;
 822                 }
 823                 /* Due to alignment rules (checked above) this cannot
 824                  * cross page boundaries
 825                  */
 826                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 827                 if (rc) {
 828                         rc = set_validity_icpt(scb_s, 0x10b0U);
 829                         goto unpin;
 830                 }
 831                 vsie_page->sdnx_gpa = gpa;
 832                 scb_s->sdnxo = hpa | sdnxc;
 833         }
 834         return 0;
 835 unpin:
 836         unpin_blocks(vcpu, vsie_page);
 837         return rc;
 838 }
 839
 840 /* unpin the scb provided by guest 2, marking it as dirty */
 841 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 842                       gpa_t gpa)
 843 {
 844         hpa_t hpa = (hpa_t) vsie_page->scb_o;
 845
 846         if (hpa)
 847                 unpin_guest_page(vcpu->kvm, gpa, hpa);
 848         vsie_page->scb_o = NULL;
 849 }
 850
 851 /*
 852  * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
 853  *
 854  * Returns: - 0 if the scb was pinned.
 855  *          - > 0 if control has to be given to guest 2
 856  */
 857 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 858                    gpa_t gpa)
 859 {
 860         hpa_t hpa;
 861         int rc;
 862
 863         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 864         if (rc) {
 865                 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 866                 WARN_ON_ONCE(rc);
 867                 return 1;
 868         }
 869         vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
 870         return 0;
 871 }
 872
 873 /*
 874  * Inject a fault into guest 2.
 875  *
 876  * Returns: - > 0 if control has to be given to guest 2
 877  *            < 0 if an error occurred during injection.
 878  */
 879 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
 880                         bool write_flag)
 881 {
 882         struct kvm_s390_pgm_info pgm = {
 883                 .code = code,
 884                 .trans_exc_code =
 885                         /* 0-51: virtual address */
 886                         (vaddr & 0xfffffffffffff000UL) |
 887                         /* 52-53: store / fetch */
 888                         (((unsigned int) !write_flag) + 1) << 10,
 889                         /* 62-63: asce id (alway primary == 0) */
 890                 .exc_access_id = 0, /* always primary */
 891                 .op_access_id = 0, /* not MVPG */
 892         };
 893         int rc;
 894
 895         if (code == PGM_PROTECTION)
 896                 pgm.trans_exc_code |= 0x4UL;
 897
 898         rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
 899         return rc ? rc : 1;
 900 }
 901
 902 /*
 903  * Handle a fault during vsie execution on a gmap shadow.
 904  *
 905  * Returns: - 0 if the fault was resolved
 906  *          - > 0 if control has to be given to guest 2
 907  *          - < 0 if an error occurred
 908  */
 909 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 910 {
 911         int rc;
 912
 913         if (current->thread.gmap_int_code == PGM_PROTECTION)
 914                 /* we can directly forward all protection exceptions */
 915                 return inject_fault(vcpu, PGM_PROTECTION,
 916                                     current->thread.gmap_addr, 1);
 917
 918         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 919                                    current->thread.gmap_addr);
 920         if (rc > 0) {
 921                 rc = inject_fault(vcpu, rc,
 922                                   current->thread.gmap_addr,
 923                                   current->thread.gmap_write_flag);
 924                 if (rc >= 0)
 925                         vsie_page->fault_addr = current->thread.gmap_addr;
 926         }
 927         return rc;
 928 }
 929
 930 /*
 931  * Retry the previous fault that required guest 2 intervention. This avoids
 932  * one superfluous SIE re-entry and direct exit.
 933  *
 934  * Will ignore any errors. The next SIE fault will do proper fault handling.
 935  */
 936 static void handle_last_fault(struct kvm_vcpu *vcpu,
 937                               struct vsie_page *vsie_page)
 938 {
 939         if (vsie_page->fault_addr)
 940                 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 941                                       vsie_page->fault_addr);
 942         vsie_page->fault_addr = 0;
 943 }
 944
 945 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
 946 {
 947         vsie_page->scb_s.icptcode = 0;
 948 }
 949
 950 /* rewind the psw and clear the vsie icpt, so we can retry execution */
 951 static void retry_vsie_icpt(struct vsie_page *vsie_page)
 952 {
 953         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 954         int ilen = insn_length(scb_s->ipa >> 8);
 955
 956         /* take care of EXECUTE instructions */
 957         if (scb_s->icptstatus & 1) {
 958                 ilen = (scb_s->icptstatus >> 4) & 0x6;
 959                 if (!ilen)
 960                         ilen = 4;
 961         }
 962         scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
 963         clear_vsie_icpt(vsie_page);
 964 }
 965
 966 /*
 967  * Try to shadow + enable the guest 2 provided facility list.
 968  * Retry instruction execution if enabled for and provided by guest 2.
 969  *
 970  * Returns: - 0 if handled (retry or guest 2 icpt)
 971  *          - > 0 if control has to be given to guest 2
 972  */
 973 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 974 {
 975         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 976         __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
 977
 978         if (fac && test_kvm_facility(vcpu->kvm, 7)) {
 979                 retry_vsie_icpt(vsie_page);
 980                 if (read_guest_real(vcpu, fac, &vsie_page->fac,
 981                                     sizeof(vsie_page->fac)))
 982                         return set_validity_icpt(scb_s, 0x1090U);
 983                 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
 984         }
 985         return 0;
 986 }
 987
 988 /*
 989  * Run the vsie on a shadow scb and a shadow gmap, without any further
 990  * sanity checks, handling SIE faults.
 991  *
 992  * Returns: - 0 everything went fine
 993  *          - > 0 if control has to be given to guest 2
 994  *          - < 0 if an error occurred
 995  */
 996 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 997         __releases(vcpu->kvm->srcu)
 998         __acquires(vcpu->kvm->srcu)
 999 {
1000         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1001         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1002         int guest_bp_isolation;
1003         int rc = 0;
1004
1005         handle_last_fault(vcpu, vsie_page);
1006
1007         if (need_resched())
1008                 schedule();
1009         if (test_cpu_flag(CIF_MCCK_PENDING))
1010                 s390_handle_mcck();
1011
1012         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1013
1014         /* save current guest state of bp isolation override */
1015         guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1016
1017         /*
1018          * The guest is running with BPBC, so we have to force it on for our
1019          * nested guest. This is done by enabling BPBC globally, so the BPBC
1020          * control in the SCB (which the nested guest can modify) is simply
1021          * ignored.
1022          */
1023         if (test_kvm_facility(vcpu->kvm, 82) &&
1024             vcpu->arch.sie_block->fpf & FPF_BPBC)
1025                 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1026
1027         local_irq_disable();
1028         guest_enter_irqoff();
1029         local_irq_enable();
1030
1031         /*
1032          * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1033          * and VCPU requests also hinder the vSIE from running and lead
1034          * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1035          * also kick the vSIE.
1036          */
1037         vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1038         barrier();
1039         if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1040                 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1041         barrier();
1042         vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1043
1044         local_irq_disable();
1045         guest_exit_irqoff();
1046         local_irq_enable();
1047
1048         /* restore guest state for bp isolation override */
1049         if (!guest_bp_isolation)
1050                 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1051
1052         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1053
1054         if (rc == -EINTR) {
1055                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1056                 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1057                 return 0;
1058         }
1059
1060         if (rc > 0)
1061                 rc = 0; /* we could still have an icpt */
1062         else if (rc == -EFAULT)
1063                 return handle_fault(vcpu, vsie_page);
1064
1065         switch (scb_s->icptcode) {
1066         case ICPT_INST:
1067                 if (scb_s->ipa == 0xb2b0)
1068                         rc = handle_stfle(vcpu, vsie_page);
1069                 break;
1070         case ICPT_STOP:
1071                 /* stop not requested by g2 - must have been a kick */
1072                 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1073                         clear_vsie_icpt(vsie_page);
1074                 break;
1075         case ICPT_VALIDITY:
1076                 if ((scb_s->ipa & 0xf000) != 0xf000)
1077                         scb_s->ipa += 0x1000;
1078                 break;
1079         }
1080         return rc;
1081 }
1082
1083 static void release_gmap_shadow(struct vsie_page *vsie_page)
1084 {
1085         if (vsie_page->gmap)
1086                 gmap_put(vsie_page->gmap);
1087         WRITE_ONCE(vsie_page->gmap, NULL);
1088         prefix_unmapped(vsie_page);
1089 }
1090
1091 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1092                                struct vsie_page *vsie_page)
1093 {
1094         unsigned long asce;
1095         union ctlreg0 cr0;
1096         struct gmap *gmap;
1097         int edat;
1098
1099         asce = vcpu->arch.sie_block->gcr[1];
1100         cr0.val = vcpu->arch.sie_block->gcr[0];
1101         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1102         edat += edat && test_kvm_facility(vcpu->kvm, 78);
1103
1104         /*
1105          * ASCE or EDAT could have changed since last icpt, or the gmap
1106          * we're holding has been unshadowed. If the gmap is still valid,
1107          * we can safely reuse it.
1108          */
1109         if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
1110                 return 0;
1111
1112         /* release the old shadow - if any, and mark the prefix as unmapped */
1113         release_gmap_shadow(vsie_page);
1114         gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1115         if (IS_ERR(gmap))
1116                 return PTR_ERR(gmap);
1117         gmap->private = vcpu->kvm;
1118         WRITE_ONCE(vsie_page->gmap, gmap);
1119         return 0;
1120 }
1121
1122 /*
1123  * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1124  */
1125 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1126                                 struct vsie_page *vsie_page)
1127 {
1128         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1129
1130         WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1131         /*
1132          * External calls have to lead to a kick of the vcpu and
1133          * therefore the vsie -> Simulate Wait state.
1134          */
1135         kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1136         /*
1137          * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1138          * automatically be adjusted on tod clock changes via kvm_sync_clock.
1139          */
1140         preempt_disable();
1141         scb_s->epoch += vcpu->kvm->arch.epoch;
1142
1143         if (scb_s->ecd & ECD_MEF) {
1144                 scb_s->epdx += vcpu->kvm->arch.epdx;
1145                 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1146                         scb_s->epdx += 1;
1147         }
1148
1149         preempt_enable();
1150 }
1151
1152 /*
1153  * Unregister a shadow scb from a VCPU.
1154  */
1155 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1156 {
1157         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1158         WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1159 }
1160
1161 /*
1162  * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1163  * prefix pages and faults.
1164  *
1165  * Returns: - 0 if no errors occurred
1166  *          - > 0 if control has to be given to guest 2
1167  *          - -ENOMEM if out of memory
1168  */
1169 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1170 {
1171         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1172         int rc = 0;
1173
1174         while (1) {
1175                 rc = acquire_gmap_shadow(vcpu, vsie_page);
1176                 if (!rc)
1177                         rc = map_prefix(vcpu, vsie_page);
1178                 if (!rc) {
1179                         gmap_enable(vsie_page->gmap);
1180                         update_intervention_requests(vsie_page);
1181                         rc = do_vsie_run(vcpu, vsie_page);
1182                         gmap_enable(vcpu->arch.gmap);
1183                 }
1184                 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1185
1186                 if (rc == -EAGAIN)
1187                         rc = 0;
1188                 if (rc || scb_s->icptcode || signal_pending(current) ||
1189                     kvm_s390_vcpu_has_irq(vcpu, 0) ||
1190                     kvm_s390_vcpu_sie_inhibited(vcpu))
1191                         break;
1192         }
1193
1194         if (rc == -EFAULT) {
1195                 /*
1196                  * Addressing exceptions are always presentes as intercepts.
1197                  * As addressing exceptions are suppressing and our guest 3 PSW
1198                  * points at the responsible instruction, we have to
1199                  * forward the PSW and set the ilc. If we can't read guest 3
1200                  * instruction, we can use an arbitrary ilc. Let's always use
1201                  * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1202                  * memory. (we could also fake the shadow so the hardware
1203                  * handles it).
1204                  */
1205                 scb_s->icptcode = ICPT_PROGI;
1206                 scb_s->iprcc = PGM_ADDRESSING;
1207                 scb_s->pgmilc = 4;
1208                 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1209                 rc = 1;
1210         }
1211         return rc;
1212 }
1213
1214 /*
1215  * Get or create a vsie page for a scb address.
1216  *
1217  * Returns: - address of a vsie page (cached or new one)
1218  *          - NULL if the same scb address is already used by another VCPU
1219  *          - ERR_PTR(-ENOMEM) if out of memory
1220  */
1221 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1222 {
1223         struct vsie_page *vsie_page;
1224         struct page *page;
1225         int nr_vcpus;
1226
1227         rcu_read_lock();
1228         page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1229         rcu_read_unlock();
1230         if (page) {
1231                 if (page_ref_inc_return(page) == 2)
1232                         return page_to_virt(page);
1233                 page_ref_dec(page);
1234         }
1235
1236         /*
1237          * We want at least #online_vcpus shadows, so every VCPU can execute
1238          * the VSIE in parallel.
1239          */
1240         nr_vcpus = atomic_read(&kvm->online_vcpus);
1241
1242         mutex_lock(&kvm->arch.vsie.mutex);
1243         if (kvm->arch.vsie.page_count < nr_vcpus) {
1244                 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
1245                 if (!page) {
1246                         mutex_unlock(&kvm->arch.vsie.mutex);
1247                         return ERR_PTR(-ENOMEM);
1248                 }
1249                 page_ref_inc(page);
1250                 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1251                 kvm->arch.vsie.page_count++;
1252         } else {
1253                 /* reuse an existing entry that belongs to nobody */
1254                 while (true) {
1255                         page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1256                         if (page_ref_inc_return(page) == 2)
1257                                 break;
1258                         page_ref_dec(page);
1259                         kvm->arch.vsie.next++;
1260                         kvm->arch.vsie.next %= nr_vcpus;
1261                 }
1262                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1263         }
1264         page->index = addr;
1265         /* double use of the same address */
1266         if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1267                 page_ref_dec(page);
1268                 mutex_unlock(&kvm->arch.vsie.mutex);
1269                 return NULL;
1270         }
1271         mutex_unlock(&kvm->arch.vsie.mutex);
1272
1273         vsie_page = page_to_virt(page);
1274         memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1275         release_gmap_shadow(vsie_page);
1276         vsie_page->fault_addr = 0;
1277         vsie_page->scb_s.ihcpu = 0xffffU;
1278         return vsie_page;
1279 }
1280
1281 /* put a vsie page acquired via get_vsie_page */
1282 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1283 {
1284         struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1285
1286         page_ref_dec(page);
1287 }
1288
1289 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1290 {
1291         struct vsie_page *vsie_page;
1292         unsigned long scb_addr;
1293         int rc;
1294
1295         vcpu->stat.instruction_sie++;
1296         if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1297                 return -EOPNOTSUPP;
1298         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1299                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1300
1301         BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1302         scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1303
1304         /* 512 byte alignment */
1305         if (unlikely(scb_addr & 0x1ffUL))
1306                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1307
1308         if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1309             kvm_s390_vcpu_sie_inhibited(vcpu))
1310                 return 0;
1311
1312         vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1313         if (IS_ERR(vsie_page))
1314                 return PTR_ERR(vsie_page);
1315         else if (!vsie_page)
1316                 /* double use of sie control block - simply do nothing */
1317                 return 0;
1318
1319         rc = pin_scb(vcpu, vsie_page, scb_addr);
1320         if (rc)
1321                 goto out_put;
1322         rc = shadow_scb(vcpu, vsie_page);
1323         if (rc)
1324                 goto out_unpin_scb;
1325         rc = pin_blocks(vcpu, vsie_page);
1326         if (rc)
1327                 goto out_unshadow;
1328         register_shadow_scb(vcpu, vsie_page);
1329         rc = vsie_run(vcpu, vsie_page);
1330         unregister_shadow_scb(vcpu);
1331         unpin_blocks(vcpu, vsie_page);
1332 out_unshadow:
1333         unshadow_scb(vcpu, vsie_page);
1334 out_unpin_scb:
1335         unpin_scb(vcpu, vsie_page, scb_addr);
1336 out_put:
1337         put_vsie_page(vcpu->kvm, vsie_page);
1338
1339         return rc < 0 ? rc : 0;
1340 }
1341
1342 /* Init the vsie data structures. To be called when a vm is initialized. */
1343 void kvm_s390_vsie_init(struct kvm *kvm)
1344 {
1345         mutex_init(&kvm->arch.vsie.mutex);
1346         INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1347 }
1348
1349 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1350 void kvm_s390_vsie_destroy(struct kvm *kvm)
1351 {
1352         struct vsie_page *vsie_page;
1353         struct page *page;
1354         int i;
1355
1356         mutex_lock(&kvm->arch.vsie.mutex);
1357         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1358                 page = kvm->arch.vsie.pages[i];
1359                 kvm->arch.vsie.pages[i] = NULL;
1360                 vsie_page = page_to_virt(page);
1361                 release_gmap_shadow(vsie_page);
1362                 /* free the radix tree entry */
1363                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1364                 __free_page(page);
1365         }
1366         kvm->arch.vsie.page_count = 0;
1367         mutex_unlock(&kvm->arch.vsie.mutex);
1368 }
1369
1370 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1371 {
1372         struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1373
1374         /*
1375          * Even if the VCPU lets go of the shadow sie block reference, it is
1376          * still valid in the cache. So we can safely kick it.
1377          */
1378         if (scb) {
1379                 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1380                 if (scb->prog0c & PROG_IN_SIE)
1381                         atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1382         }
1383 }